BACKGROUND OF THE INVENTION
[0001] The present invention relates to a method and an apparatus for canceling echoes in
a system using plural received signals and a single or plural transmission signals.
[0002] There are multi-channel echo canceling methods or apparatuses that remove echoes
caused by received signals propagating through spatial acoustic paths in the system
of transmitting plural received signals or a single or plural transmission signals.
Two types of multi-channel echo canceling apparatuses including a cascade connection
type and a linear coupled type have been proposed in the Technical Report of the Institute
of Electronics, Information and Communication Engineers (IEICE) of Japan, Vol. 84,
No. 330, pp. 7-14, CS-84-178 (hereinafter referred to as reference 1). According to
reference (1), the cascade connection type is inferior to the linear coupled type
in echo suppression capability caused by constitutional constraint. Here, the case
where a linear coupled multi-channel echo canceling apparatus (echo canceller) is
applied to a two-channel system, together with received signals and transmission signals,
will be described below as a prior art.
[0003] Fig. 24 shows a linear coupled multi-channel echo canceller. The first loudspeaker
3 reproduces the first received signal 1. The second loudspeaker 4 reproduces the
second received signal 2. The first echo 5 which is generated as the signal propagating
via spatial acoustic paths from the first loudspeaker 3 to the first microphone 9
and the second echo 6 which is produced as the signal propagating through spatial
acoustic paths from the second loudspeaker 4 to the first microphone 9 are added to
the first transmission signal 12 or the voice of the talker 11 received with the first
microphone 9, so that a first mixed signal 14 is created. Similarly, the third echo
7 which is generated as the signal propagating the spatial acoustic paths from the
first loudspeaker 3 to the second microphone 10 and the fourth echo 8 which is produced
as the signal propagating through spatial acoustic paths from the second loudspeaker
4 to the second microphone 10 are added to the second transmission signal 13 or the
voice of the talker 11 received with the second microphone 10, so that a second mixed
signal 15 is created.
[0004] In order to remove the echoes introduced into the first mixed signal 14, the first
adaptive filter 121 receives the first received signal 1 and then creates an echo
replica 125 corresponding to the first echo 5. The second adaptive filter 122 receives
the second received signal 2 and then creates an echo replica 126 corresponding to
the second echo 6. The first subtracter 129 subtracts the echo replicas 125 and 126
corresponding to the first echo 5 and the second echo 6, respectively, from the first
mixed signal 14. The first and second adaptive filters 121 and 122 are controlled
to minimize the output of the first subtracter 129. The output of the first subtracter
129 becomes the first output signal 16 of the echo canceller 100.
[0005] In order to remove the echoes introduced into the second mixed signal 15, the third
adaptive filter 123 receives the first received signal 1 and then creates an echo
replica 127 corresponding to the third echo 7. The fourth adaptive filter 124 receives
the second received signal 2 and then creates an echo replica 128 corresponding to
the fourth echo 8. The second subtracter 130 subtracts the replicas 127 and 128 corresponding
to the third echo 7 and the fourth echo 8 from the second mixed signal 15, respectively.
The third and fourth adaptive filters 123 and 124 are controlled to minimize the output
of the second subtracter 130. The output of the second subtracter 130 becomes the
second output signal 17 of the echo canceller 100.
[0006] In the multi-channel teleconferencing system being one of important applications
for multi-channel echo cancellation, plural microphones receive voices of talkers.
Hence, received signals by each microphone can be approximately regarded as signals
with an attenuation and a delay each depending on the distance between the talker
and the microphone. The mutual correlation between received signals in different channels
becomes very high. Let us now assume that the second received signal 2 was generated
as a delayed version of the first received signal 1, an echo path can be modeled as
a finite impulse response filter, and an echo canceller uses linear combination with
adaptive finite impulse response filters.
[0007] It is now assumed that x
1(n) is a first received signal 1 at the time n, x
2(n) is a second received signal 2 at the time n, and d(n) is an echo introduced into
the mixed signal 14. When the time differences between the first received signals
1 and the second received signals 2 is n
d (natural number), X
2(n) is expressed by:

[0008] For simplification, let us now assume that all the spatial acoustic paths ranging
from the first and second loudspeakers 3 and 4 to the first and second microphones
9 and 10 are equal in the impulse response length (N). The echo d(n) mixed in the
mixed signal 14 is the sum of the echo 5 and the echo 6 and is expressed by:

where h
1,j is an impulse response sample value of a spatial acoustic path ranging from the loudspeaker
3 to the microphone 9, h
2,i is an impulse response sample value of a spatial acoustic path ranging from the loudspeaker
4 to the microphone 9, and i is an integer value between 0 and N-1.
[0009] When x
2(n) is eliminated by substituting the equation (1) into (2), the following equation
(3) is obtained:

[0010] The echo replica d(n) hat created by each of the adaptive filters 121 and 122 is
expressed by:

where w
1,i(n) is the i-th filter coefficient of the adaptive filter 121 and w
2,i(n) is the i-th filter coefficient of the adaptive filter 122.
[0011] When x
2(n) is eliminated by substituting the equation (1) into the equation (4), the following
equation (5) is obtained:

The residual echo e(n) is expressed by:

[0012] The condition of completely eliminating echoes from the equation (6) is expressed
by:

[0013] The following expression is uniquely determined by equation (7).

However, it is clear that the following expression has an unlimited number of solutions.

[0014] Particularly the solutions of the following equation depend on values of n
d.

That is, a change of n
d caused by a shift of the talker' s position leads to a change of the solution. This
means that the echo canceling capability is degraded even with no variations in echo
path. This creates an obstacle to applications in actual environments. In these discussions,
the adaptive filters 121 and 122 used to remove echoes mixed in the mixed signal 14
have been assumed. Similarly, this discussion is applicable to adaptive filters 123
and 124.
[0015] In order to solve the above-mentioned problems, there is a multi-channel echo canceling
method where adaptive filters, respectively corresponding one to one to mixed signals,
create echo replicas in response to one received signal. Thus, one adaptive filter
per channel can estimate echoes caused by signals propagating from a single sound
source in plural paths. This method was proposed in the IEEE Proceedings of International
Conference on Acoustics, Speech and Signal Processing, Vol. 2, 1994, pp. 245-248 (hereinafter
referred to as reference 2).
[0016] The multi-channel echo canceling method of the reference 2 can avoid indefinite solutions
because one adaptive filter removes the echo generated in one channel. As a result,
the adaptive filter coefficients are converged to optimum values which are uniquely
determined. However, the reference 2 discloses evaluation results proving the fact
that the echo suppression is degraded when the parameters determined by the environment
in use such as the locations of microphones receiving the taker s voice are not within
a certain range. Hence, a multi-channel echo canceller based on linear combination
has to be used, by considering the use in a wide variety of environments.
[0017] Based on the above-mentioned premise, there is a method of uniquely determining the
adaptive filter coefficients, for a multi-channel echo canceller based on linear combination,
where the received signal is delayed and then is alternately selected as a new received
signal the delayed signal in place of the received signal. This method was proposed
in Technical Report, the Institute of Electronics, Information and Communication Engineers
of Japan, Vol. DSP97-1, 1997, pp. 1-8 (hereinafter referred to as reference 3). In
this multi-channel echo canceling method, the number of conditions used to calculate
adaptive filter coefficients is increased because of the introduction of the delayed
received signals, therefore, indefinite solutions do not occur. As a result, the adaptive
filter coefficients converge to optimum values which are uniquely determined. However,
shift of the acoustic image may often be perceived when the received signal is switched
between the original and its delayed version. In order to overcome such a problem,
there is a method of correcting the magnitudes of signals in both channels when the
received signal and the delayed received signal are switched. This method was proposed
in the Proceedings of the 12-th digital signal processing symposium of the Institute
of Electronics, Information and Communication Engineers of Japan, 1997, pp. 531-536
(hereinafter referred the as reference 4). Moreover, there is a method that can uniquely
determine adaptive filter coefficients by applying non-linear processing to the received
signals in both channels, instead of switching the received signal and the delayed
received signal. This method was proposed in the Proceedings of the IEEE International
Conference on Acoustic, Speech and Signal Proceeding, Vol. 1, 1997, pp. 303-306 (hereinafter
referred to as reference 5). However, the method disclosed in reference 4 provides
a slow convergence rate, compared with the multi-channel echo canceller based on linear
combination. Moreover, the Proceeding of the IEEE International Conference on Acoustics,
Speech and Signal Processing, Vol. 6, 1998, pp. 3677-3680 (hereinafter referred to
as reference 6) discloses that the method in the reference 5 provides a slower convergence
rate, compared with the method in reference 4.
[0018] As described in detail by referring to Fig. 24, the conventional multi-channel echo
canceling method and apparatus has the disadvantage in that the solutions to adaptive
filter coefficients are indefinite and are not guaranteed to be equal to the solutions
uniquely determined by the impulse responses in the echo paths. Moreover, in the method
proposed in reference 3, a shift of an acoustic (or sound) image may often be perceived
when the received signal or the delayed received signal is selected. Moreover, in
the methods proposed in the references 4 and 5, the convergence is slow, compared
with the conventional multi-channel echo canceling method or apparatus.
SUMMARY OF THE INVENTION
[0019] The objective of the invention is to provide a multi-channel echo canceling method
and/or apparatus with an excellent sound quality and a short convergence time, wherein
the coefficient values of the adaptive filters are converged to correct values which
are uniquely determined by the impulse responses of the echo paths.
[0020] In the multi-channel echo canceling apparatus according to the present invention,
received signals are non-linearly processed while one obtained by delaying a received
signal is used as a new received signal.
[0021] Specifically, the multi-channel echo canceling apparatus has magnitude processing
circuits for non-linearly processing received signals and a delay processing circuit
for delaying the received signal 2 and then supplying the delayed signal to adaptive
filters and to a digital-to-analog converter (DAC).
[0022] Furthermore, the multi-channel echo canceling apparatus non-linearly processes the
received signals. The multi-channel echo canceling apparatus also uses a delayed received
signal as a new received signal and applies magnitude correction to the other input
signals.
[0023] Specifically, the multi-channel echo canceling apparatus has a magnitude processing
circuit for non-linearly processing the received signal, a delay processing circuit
for delaying a non-linearly processed signal and then supplying the delayed signal
to the adaptive filters 122 and 124 and the digital-to-analog converter, a magnitude
processing circuit for non-linearly processing the received signal, and a magnitude
compensating circuit for correcting the magnitude of the non-linearly processed signal
and then supplying the corrected signal to the adaptive filters and the digital-to-analog
converter.
[0024] In the multi-channel echo canceling apparatus and or method according to the present
invention, one received signal is filtered to create a supplement signal. The adaptive
filter is operated using a new signal obtained by time-multiplexing an original received
signal and the created supplement signal. A plurality of adaptive filters estimate
echoes caused by propagating from one signal source in plural paths. For that reason,
an increase in the number of conditions for solving adaptive filter coefficients makes
solutions definite eliminating the existing problem of indefinite coefficients. Hence,
the adaptive filter coefficients are converged to optimum values which are uniquely
determined. Moreover, the multiplex parameters of the original received signal and
the supplement signal are controlled based on the nature of the received signal while
movement of the sound image caused by introducing the supplement signal is canceled
in the magnitude correction process to the input signal. Hence, degradation of the
sound quality of the audible received signal directly supplied to the loudspeaker
is suppressed so that a good sound quality can be maintained. Moreover, the synergistic
effect of introducing the non-linear processing and the supplement signal can shorten
the convergence time.
BRIEF DESCRIPTION OF THE DRAWINGS
[0025] These and other objects, features and advantages of the present invention will become
more apparent upon a reading of the following detailed description and drawings, in
which:
Fig. 1 is a block diagram of a multi-channel echo canceller according to a first embodiment
of the present invention;
Fig. 2 is a block diagram illustrating the configuration of the magnitude processing
circuit 605 or 606;
Fig. 3 is a block diagram illustrating a first configuration of the delay processing
circuit 200;
Fig. 4A and 4B are block diagrams illustrating the configuration of the filter 213;
Fig. 5 is a block diagram illustrating the second configuration of the delay processing
circuit 200;
Fig. 6 is a block diagram illustrating the analysis circuit 221 according to a first
embodiment of the present invention;
Fig. 7 is a block diagram illustrating the analysis circuit 221 according to a second
embodiment of the present invention;
Fig. 8 is a block diagram illustrating the third configuration of the delay processing
circuit 200;
Fig. 9 is a block diagram illustrating the fourth configuration of the delay processing
circuit 200;
Fig. 10 is a graph plotting the time-varying of the coefficient co(k) of the coefficient multiplier of the filter shown in Fig. 3;
Fig. 11 is a block diagram illustrating the fifth configuration of the delay processing
circuit 200;
Fig. 12 is a block diagram illustrating the sixth configuration of the delay processing
circuit 200;
Fig. 13 is a block diagram illustrating a multi-channel echo canceller according to
a second embodiment of the present invention;
Fig. 14 is a block diagram illustrating a multi-channel echo canceller according to
a third embodiment of the present invention;
Fig. 15 is a block diagram illustrating the configuration of the magnitude processing
circuit 607 or 608;
Fig. 16 is a block diagram illustrating a multi-channel echo canceller according to
a fourth embodiment of the present invention;
Fig. 17 is a block diagram illustrating a multi-channel echo canceller according to
a fifth embodiment of the present invention;
Fig. 18 is a block diagram illustrating the configuration of the filter 213 or 230
in the delay processing circuit 300;
Fig. 19 is a block diagram illustrating the first configuration of the filter 213
or 230 in the magnitude compensating circuit 400;
Fig. 20 is a block diagram illustrating the second configuration of the filter 213
or 230 in the magnitude compensating circuit 400;
Fig. 21 is a block diagram illustrating a multi-channel echo canceller according to
a sixth embodiment of the present invention;
Fig. 22 is a block diagram illustrating a multi-channel echo canceller according to
a seventh embodiment of the present invention;
Fig. 23 is a block diagram illustrating a multi-channel echo canceller according to
an eighth embodiment of the present invention; and
Fig. 24 is a block diagram illustrating a multi-channel echo canceller based on Linear
Combination.
DESCRIPTION OF THE EMBODIMENTS
[0026] Embodiments of the present invention will be described in detail below with reference
to Figs. 1 to 23. It is now assumed to use a two-channel acoustic echo canceller that
has a first received signal, a second received signal, a first mixed signal and a
second mixed signal. This acoustic echo canceller can remove acoustic echoes caused
by received signals propagating from a loudspeaker to a microphone via a spatial acoustic
path.
[0027] Fig. 1 shows a multi-channel echo canceller with two received signals and two transmission
signals, according to an embodiment of the present invention. This multi-channel echo
canceller differs from the multi-channel echo canceller based on linear combination
of Fig. 24 in that the non-linear processing circuit 600 and the pre-processing circuit
500 process the received signal 2 supplied to the adaptive filters 122 and 124. The
first mixed signal 14 and the second mixed signal 15 are created in a similar manner
to that of the multi-channel echo canceller based on linear combination shown in Fig.
24. In the non-linear processing circuit 600, the magnitude processing circuit 605
non-linearly processes the received signal 2 while the magnitude processing circuit
606 non-linearly processes the received signal 1. The first and the second received
signals 1 and 2, which are non-linearly processed, are transferred to the processing
circuit 500. If there is a slight difference between the received signals 1 and 2,
the non-linear process enlarges it, thus, reducing the correlation between the received
signals in these channels. This means that the adaptive filter coefficients are properly
identified. However, as described earlier, the non-linear processing alone results
in a slow convergence rate. For that reason, it is intended to further reduce the
correlation between channels by additionally using the pre-processing circuit 500.
The delay processing circuit 200 in the pre-processing circuit 500 processes the output
of the non-linear processing circuit 600 and then transmits the delayed process signal
to the adaptive filters 122 and 124 and the digital-to-analog converter (DAC) 19.
[0028] Fig. 2 is a block diagram illustrating the configuration of the magnitude processing
circuit 605 or 606. The magnitude processing circuit 606 consists of the coefficient
multiplier 691 and the polarity detection circuit 692 each receiving the received
signal 1. The coefficient multiplier 691 multiplies the input signal, or the received
signal 1, by α. The polarity detection circuit 692 outputs 1 when the polarity of
the input signal is positive and 0 when the polarity of the input signal is negative.
The multiplier 693 receives the output of the coefficient multiplier 691 and the output
of the polarity detection circuit 692, and outputs the product to the adder 694. The
adder 694 has the other input terminal receiving the received signal 1 without any
change. That is, with respect to the n-th received signal sample x
1(n), the adder 694 outputs either (1+α)x
1(n) when the polarity of the input signal is positive or x
1(n) when the polarity of the input signal is negative. Similarly, the magnitude processing
circuit 605 outputs either (1+α)x
2(n) when the polarity of the input signal is positive or x
2(n) when the polarity of the input signal is negative.
[0029] Fig. 3 is a block diagram illustrating the configuration of the delay processing
circuit 200. The magnitude processing circuit 605 inputs its output signal to one
input terminal of the switch 210 and the filter 213 via the input terminal 201. The
filter 213 filters the signal from the magnitude processing circuit 605 and then inputs
the filtered signal to the other input terminal of the switch 210. In other words,
the signal from the magnitude processing circuit 605 and the signal obtained by filtering
the signal from the magnitude processing circuit 605 with the filter 213 are respectively
supplied to the two input terminals of the switch 210. The frequency divider 212 inputs
a control signal to the switch 210. The control signal is created by frequency-dividing
the clock signal from the clock signal generator 211. Rectangular pulses with the
period equal to the sampling period T of the received signal 2 are generated as the
clock signal. Only for convenience in explanation, the clock signal generator 21 is
shown in Fig. 3. Actually, it is rare to have a dedicated clock signal generator in
the delay processing circuit 200. In this case, a common clock signal to the whole
system is supplied to the frequency divider 212 from outside of the delay processing
circuit 200. If the frequency divider 212 is a 1/M frequency divider for dividing
the frequency of the input signal with a factor of 1/M, the frequency divider 212
alternately outputs a level of "1" and a level of "0" to the switch 210 with a period
of MT/2. The switch 210 selects the signal from the magnitude processing circuit 605
or the output signal from the filter 213 in synchronous with the leading edge of a
rectangular pulse from the frequency divider 212 to transfer it to the output terminal
202. The signal delayed by the above-mentioned procedure is output as a pre-processed
signal from the output terminal 202.
[0030] Fig. 4A is a block diagram illustrating the configuration of the filter 213. Here,
an L-tap FIR (Finite Impulse Response) filter is assumed as the filter 213. However,
another type of a filter such as an IIR filter may also be used as the filter 213.
Referring to Fig. 4A, the intput terminal 2130 receives the output signal from the
magnitude processing circuit 605 of Fig. 1. The signal of the output terminal 2134
is supplied to the switch 210 of Fig. 3. The signal of the input terminal 2130 is
input to the delay element 2131
1 and the coefficient multiplier 2132
0. Each of delay elements 2131
1, 2131
2,···, 2131
L-1 is a unit delay element for delaying the input signal by one sample. The delay elements
are connected in a cascade form to construct a tapped delay line with L taps. The
coefficient multipliers 2132
0, 2132
1,···, 2132
L-1 have coefficients c
0, c
1, ···, c
L-1, respectively. For L = 2, c
0 = 0 and c
1 = 1, the filter 213, for example, consists of only the delay element 2131
1, as shown in Fig. 4B. For M = 1, that is, for the case where the frequency divider
212 of Fig. 1 does not divide the frequency, the embodiment of the present invention
shown in Fig. 4B is equivalent to the method disclosed in the Technical Report of
the Institute of Electronics, Information and Communication Engineers of Japan, Vol.
DSP96-100, 1996, pp. 17-24 (reference 7). The reference 7 analytically shows that
the adaptive filter coefficients are uniquely determined.
[0031] Let us now consider M > 1. In this case, it is apparent that the number of conditionals
for calculating the adaptive filter coefficients is the same as that for M = 1. Consequently,
the adaptive filter coefficients can be uniquely determined. Even general cases, where
the filter 213 cannot be expressed by L = 2, c
0 = 0 and c
1 = 1, can be handled in a similar manner. When the output of the filter 213 is equivalent
to the input signal, that is, the filter 213 is expressed by L = 1 and c
0 = 1, the output of the delay processing circuit 200 depends on the status of the
switch 210. Consequently, the number of conditionals for calculating adaptive filter
coefficients is equal to the case where the filter 213 is expressed by L = 2, c
0 = 0 and c

= 1. Thus, the adaptive filter coefficients can be uniquely determined.
[0032] Except for M = infinite, discontinuity occurs in the output of the switch 210 caused
by the switching operation of the switch 210. This signal discontinuity is subjectively
perceived as noise. The frequency of the noise is inversely proportional to the value
of M. As M has a large value, it is more difficult to perceive the signal discontinuity,
compared with M of a small value, however, the perception itself is unavoidable. The
reference 3 discloses that the noise can be suppressed by time-varying coefficients
c
j (where j = 0, 1, ···, L-1) of the filter 213, together with an example of how to
control the time-varying coefficients.
[0033] As the coefficient update algorithm for the adaptive filters 121, 122, 123 and 124,
the LMS algorithm is disclosed in the "Adaptive Signal Processing" , 1985, pp. 99-113,
Prentice-Hall Inc., USA (hereinafter referred to as reference 8) and the normalized
LMS (NLMS) algorithm is disclosed in "Adaptive Filters" , 1985, pp. 49-56, Kulwer
Academic Publications, USA (hereinafter referred to as reference 9). It is now assumed
that the adaptive filters 121 and 122 are controlled using the LMS algorithm with
the same step sizes. The i-th coefficient of the adaptive filter 121 w
1,i(n+1) after the (n+1)-th adaptation is expressed by its value w
1,i(n) after the n-th adaptation, as follows:

[0034] The i-th coefficient of the adaptive filter 122 w
2,i(n+1) after the (n+1)-th adaptation is expressed by its value w
2,i(n) after the n-th adaptation, as follows:

[0035] This is applicable to the coefficient adaptation of the adaptive filters 123 and
124.
[0036] Fig. 5 is a block diagram illustrating the delay processing circuit 200 with the
second configuration. The second configuration differs from the first configuration
of Fig. 3 in that the delay processing circuit 200 has the analysis circuit 221 and
the AND circuit 220, in addition to the frequency divider 212. In the first configuration
of Fig. 3, the switch 210 changes its status every M samples. However, in the configuration
of Fig. 5, the switch 210 changes its status according to the logical product of the
output signal of the frequency divider 212 and the output signal from the analysis
circuit 221. The analysis circuit 221 analyzes the output signal from the magnitude
processing circuit 605. The analysis circuit 221 produces 1 when the analyzed result
satisfies a predetermined condition and produces "0" when the analyzed result does
not satisfy the predetermined condition, then, outputting it to the AND circuit 220.
The frequency divider 212 supplies a control signal "0" or "1" to the AND circuit
220. The AND circuit 220 detects that the outputs from the analysis circuit 221 and
the frequency divider 212 are identical, i.e. the time information coincides with
a period of M samples and the analysis results of the input satisfy the predetermined
condition simultaneously, and then outputs the signal to control the switching of
the switch 210.
[0037] Various methods may be considered to analyze the output of the magnitude processing
circuit 605 with the analysis circuit 221. For example, upon focusing on the suppression
of subjective noises due to signal discontinuity, variations in magnitude of the output
signal of the magnitude processing circuit 605 may be detected.
[0038] Fig. 6 shows the analysis circuit according to the first embodiment.
[0039] Referring to Fig. 6, the analysis circuit 221 consists of a delay element 2210, a
subtracter 2211, an absolute value circuit 2212, a decision circuit 2213, and a memory
2214. The delay element 2210 and the subtracter 2211 receive the output signal from
the magnitude processing circuit 605, that is, the input signal to the analysis circuit
221. The delay element 2210 delays the input signal by one sample period and then
outputs the delayed signal to the subtracter 2211. The subtracter 2211 subtracts the
output signal of the delay element 2210 from the output signal of the magnitude processing
circuit 605 and then outputs the difference to the absolute value circuit 2212. The
absolute value circuit 2212 obtains the absolute value of its input and outputs the
result to the decision circuit 2213. The memory 2214 supplies a threshold value θ
to the decision circuit 2213. The decision circuit 2213 outputs "1" when the signal
from the absolute value circuit 2212 is smaller than the threshold value and otherwise
outputs "0" . The decision circuit 2213 outputs its output to the AND circuit 220
of Fig. 5. Namely, when variations in magnitude of the output signal of the magnitude
processing circuit 605 are small, the switch 210 changes its status.
[0040] Fig. 7 shows the analysis circuit 221 based on a post masking, according to the second
embodiment. The post masking is a phenomenon in which signal samples with a small
magnitude following a certain signal sample cannot be perceived by a human' s ear.
Post masking is described in detail in "Pcychoacoustics" by Zwicker, translated by
Yamada, published by Nishmura Syoten, 1992, pp. 132-146 (reference 10). Referring
to Fig. 7, the analysis circuit 221 consists of a group of delay elements 2215
0, 2215
1, ···, 2215
N-1, a group of difference evaluation circuits 2216
0, 2216
1, ···, 2216
N-1, and a control signal generator 2217. N is a positive integer. The magnitude processing
circuit 605 outputs its output signal to the delay element 2215
0 and the difference evaluation circuit 2216
0. Delay elements 2215
0, 2215
1, ···, 2215
N-1 form a tapped delay line which delays the supplied signal by one sample period.
[0041] The difference evaluation circuit 2216
0 evaluates the difference between the output signal from the magnitude processing
circuit 605 and the signal supplied from the delay element 2215
0 and outputs the result to the control signal generator 2217. Difference evaluation
can be performed by subtracting the output signal from the magnitude processing circuit
605 from the signal supplied from the delay element 2215
0.
[0042] Then, "1" is output when the result is larger than a predetermined threshold value
δ
0.
[0043] "0" is output when the result is smaller than the predetermined threshold value δ
0. Moreover, by subtracting the absolute value of an output signal from the magnitude
processing circuit 605 from the absolute value of a signal supplied from the delay
element 2215
0, "1" is output when the result is larger than a predetermined threshold value ε
0 and "0" is output when the result is smaller than a predetermined threshold value
ε
0.
[0044] Similarly, each of the difference evaluation circuits 2216
0, 2216
1, ···, 2216
N-1 evaluates the difference between the signal supplied from the corresponding delay
element and the output signal from the magnitude processing circuit 605 and then outputs
the result to the control signal generator 2217. The control signal generator 2217
creates a control signal using evaluation results of difference supplied from the
difference evaluation circuits. The control signal, for example, can be created based
on a coincidence among the input signals from the difference evaluation circuits.
That is, "1" is output for a coincidence while "0" is output for a non-coincidence.
The majority of signals from the differential evaluation circuits may be used as a
control signal. When "1" holds the majority, "1" is output. When "1" holds a minority,
"0" is output. As another example, respective input signals may be multiplied by a
predetermined independent constant. The sum of the multiplication results is compared
with a predetermined threshold. When the sum is larger than the threshold value, "1"
is output. Otherwise, "0" is output. It is apparent that the control signal generator
2217 may use a coincidence between respective multiplication results or their majority.
In the above-mentioned procedure, when the magnitude of the output from the magnitude
processing circuit 605 decreases below the past value, the switch changes its state.
[0045] The reference 10 discloses pre-masking as a similar phenomenon to post masking. Pre-masking
is a phenomenon in which signals with a small magnitude are masked by signals consecutive
thereto, thus, are inaudible by a human' s ear. In order to detect pre-masking, the
whole samples of the signal must be delayed. That is, delay elements must be inserted
in the two input paths for the switch 210 shown in Fig. 5. In association with the
insertion of delay elements, delay elements with a corresponding delay amount are
inserted in the path for the received signal 1 before the adaptive filters 121 and
123 to adjust the delay of the received signal 1. The delay amounts of the delay elements
depend on the time required for pre-masking detection. For example, at least a 2-sample
delay is required to detect the pre-masking due to the signal delayed by 2 samples.
In the calculation by the difference evaluation circuits 2216
0, 2216
1, ···, 2216
N-1 shown in Fig. 7, the outputs must be inverted. That is, "0" is output when "1" is
to be originally output while "1" is output when "0" is to be originally output. This
inversion allows pre-masking to be detected. In the above-mentioned procedure, the
switch 210 changes immediately its state before the magnitude of the output signal
of the magnitude processing circuit 605 increases.
[0046] In the configuration of Fig. 5, at least for M sample, after the output signal of
the frequency divider 212 does not match with the output of the analysis circuit 221,
the switch 210 cannot change its state. That is, the switching period of the switch
210 is limited to an integer multiple of M. The delay processing circuit 200 may be
constructed such that the switching period of the switch 210 is not limited to an
integer multiple of M.
[0047] Fig. 8 is a block diagram illustrating the delay processing circuit 200 with the
third configuration. This configuration differs from the second configuration of Fig.
5 in that a new analysis circuit 222 is added instead of the frequency divider 212,
the analysis circuit 221 and the AND circuit 220. In the configuration of Fig. 5,
the switch 210 is controlled by the logical AND between the output of the frequency
divider 212 and the output of the analysis circuit 221. However, in the configuration
of Fig. 8, the analysis circuit 222 analyzes the output signal of the magnitude processing
circuit 605 and then directly creates a control signal for the switch 210 using the
analysis result and rectangular pulses. The rectangular pulses are supplied from the
clock generation circuit 211 to the analysis circuit 222.
[0048] The analysis circuit 222 performs basically the same analysis as the analysis circuit
221. The analysis circuit 222 can detect a variance in magnitude of the output signal
from the magnitude processing circuit 605 or can perform analysis based on the pre/post
masking. When the analysis result indicates that the output signal of the magnitude
processing circuit 605 satisfies predetermined conditions and a predetermined sampling
period (M
2T) or more has passed from the immediate previous switching operation, the analysis
circuit 222 outputs a control signal of "1". M
2 is a positive integer satisfying M
2 > 1. In other cases, the analysis circuit 222 outputs a control signal of "0" . The
analysis circuit 222 outputs the control signal to the switch 210 to control the switching
thereof. Specifically, the sampling period is evaluated by counting the rectangular
pulses with a counter and then comparing the count with the value M
2 stored in a memory. If the count is equal to the value M
2, it is judged that the sampling period (M
2T) has passed. Thus, the analysis circuit 222 outputs "1" while the counter is reset.
In order to suppress the subjective noise due to signal discontinuity, a case with
time-varying of the coefficients c
j (where j = 0, 1, ···, L-1) of the filter 213 has been explained for L = 2, using
Fig. 3. However, the delay process circuit 200 without the switch 210 in Fig. 2, 5
or 8 may be used instead by suitably setting the coefficients c
0(k) and c
1(k).
[0049] Fig. 9 is a block diagram illustrating the delay processing circuit 200 with the
fourth configuration. The filter 230 receives the received signal 2 via the input
terminal 201. The filter 230 filters the received signal 2 and outputs the filtered
signal to the output terminal 202. The clock generation circuit 211 and the frequency
divider 212 respectively input control signals to the filter 230. The clock generation
circuit 211 generates rectangular pulses of a period equal to the sampling period
T of the received signal 2. The frequency divider 212 supplies the control signal
obtained by frequency-dividing the clock from the clock generation circuit 211. The
filter 230 controls the time-varying coefficient based on the control signals.
[0050] For L = 2 in Fig. 3, c
0(k) is shown in Fig. 10. c
1(k) is expressed by:

where i is an arbitrary natural number.
[0051] c
o(k) takes c
o(0) and 0 alternately with a period of 2MT. Strictly, only the first JT and the last
JT in a zero-valued range the value of c
0(k) directly changes from c
0(0) to 0 or from 0 to c
0(0). Since c
1(k) is given by the equation (10), either c
0(k) or c
1(k) is zero while the other is non-zero, almost over the whole period of time. That
is, c
0(k) and c
1(k) are exclusive to each other. A similar switching operation to that of the switch
210 can be executed without the switch 210 of Fig. 3. It may be considered that L
≠ 2 corresponds to taps of the filter 230 connected in parallel. In other words, c
0(k) and c
1(k), c
2(k), ···, c
L-1(k) are exclusive to each other. When one coefficient is zero, the other coefficients
become non-zero. Each of c
1(k), c
2(k), ···, c
L-1(k) and the corresponding vale of J may respectively have a different value.
[0052] Fig. 11 is a block diagram illustrating the delay processing circuit 200 with the
fifth configuration. The filter 203 receives the output signal of the magnitude processing
circuit 605 input via the input terminal 201. The filter 203 filters the output signal
from the magnitude processing circuit 605 and then outputs the filtered signal to
the output terminal 202. The filter 230 also receives the control signals from the
clock generation circuit 211 and the AND circuit 220. The AND circuit 220 receives
the signal from the analysis circuit 211 and the signal from the frequency divider
212. The clock generation circuit 211 generates rectangular pulses with a period equal
to the sampling period T of the output signal from the magnitude processing circuit
605. The frequency divider 212 supplies a control signal to the AND circuit 220. The
control signal is created by frequency-dividing the clock from the clock generation
circuit 211. The analysis circuit 221 analyzes the output of the magnitude processing
circuit 605 and then outputs "1" to the AND circuit 220 when the result of the analysis
satisfies a predetermined condition and "0" to the AND circuit 220 when the result
of the analysis does not satisfy the predetermined condition. As described earlier,
the AND circuit 220 receives a control signal of "0" or 1 from the frequency divider
212. The AND circuit 220 detects that the output of the analysis circuit 221 is identical
to the output of the frequency divider 212. That is, the time information coincides
with the M sample periods and the analysis result on the input signal satisfies the
predetermined conditions. Then, the AND circuit 220 outputs its output signal to the
filter 230. The filter 230 controls the time-varying coefficients based on the control
signals.
[0053] Fig. 12 is a block diagram illustrating the delay processing circuit 200 with the
sixth configuration. This configuration differs from the fifth configuration (described
with Fig. 1) in that a new analysis circuit 222 is added instead of the frequency
divider 212, the analysis circuit 221 and the AND circuit 220. That is, in the configuration
of Fig. 11, the time-varying coefficients of the filter 230 are controlled by the
logical AND between the output signals of the frequency divider 212 and the analysis
circuit 221. However, in the configuration of Fig. 12, the analysis circuit 222 analyzes
the output signal of the magnitude processing circuit 605. Then, the control signals
for the filter 230 are directly created using the analysis result and the rectangular
pulses supplied from the clock signal generation circuit 211 to the analysis circuit
222.
[0054] The pre-processing circuit 500, which has the delay processing circuit 200 receiving
the received signal 2, has been described by referring to Fig. 1. The pre-processing
circuit 500 may have the delay processing circuit 200 which receives the received
signal 1.
[0055] Fig. 13 shows a multi-channel echo canceller with two channels for the received signals
and two channels for the transmission signals, according to the second embodiment
of the present invention. The second embodiment differs from the first embodiment
of Fig. 1 in that the pre-processing circuit 500 and the non-linear process circuit
600 are exchanged in the order. As described earlier, the pre-processing and the non-linear
processing are independently applied. Hence, the configuration in which the order
of these processings is exchanged can provide similar effects.
[0056] Fig. 14 shows a multi-channel echo canceller with two channels for the received signals
and two channels for the transmission signals, according to the third embodiment of
the present invention. The third embodiment differs from the first embodiment of Fig.
1 in that the non-linear processing circuit 601 is substituted with the non-linear
processing circuit 600. The non-linear processing circuit 610 includes the magnitude
processing circuits 607 and 608 each having the same configuration. Hence, the difference
between the non-linear processing circuits 600 and 601 is the difference between magnitude
processing circuits 605 and 607.
[0057] Fig. 15 is a block diagram illustrating the configuration of the magnitude processing
circuit 607 or 608. This configuration differs from that of the magnitude processing
circuit 605 or 606 of Fig. 2 in that the coefficient multiplier 691 receives the received
signal 2 instead of the received signal 1. The adder 694 produces (x
1(n) + αx
2(n)) for the n-th sample of the received signal when the polarity of the n-th received
signal sample x
1(n) is positive and produces x
1(n) when the polarity of the n-th received signal sample x
1(n) is negative. The produced signal becomes the output signal of the magnitude processing
circuit 608. Similarly, the magnitude processing circuit 607 produces (x
2(n) + αx
1(n)) when the polarity of the input signal is positive and produces x
2(n) when the polarity of the input signal is negative. In this embodiment, since the
other channel is used for non-linear processing, the variation from the original signal
becomes large. Hence, reduction of the correlation between the channels can be large.
[0058] Fig. 16 shows a multi-channel echo canceller with two channels for the received signals
and two channels for the transmission signals according to the fourth embodiment of
the present invention. Like the first embodiment of Fig. 1 and the second embodiment
of Fig. 13, the pre-processing circuit 500 and the non-linear processing circuit 610
are exchanged in the order in the third embodiment of Fig. 14. Because of the same
reason as that in the second embodiment, the fourth embodiment has effects identical
to those by the third embodiment of Fig. 14.
[0059] Next, a case where a new pre-processing circuit is used, wherein the delay processing
circuit is applicable to the received signal 2 to create a delayed signal and the
magnitude compensating circuit is applicable to the received signal 1, will be described
below.
[0060] Fig. 17 shows a multi-channel echo canceller with two channels for the received signals
and two channels for the transmission signals according to the fifth embodiment of
the present invention. Unlike the first embodiment of Fig. 1, the delay processing
circuit 605 delays the received signal 2 for the adaptive filters 122 and 124 and
the magnitude compensating circuit 606 compensates the amplitude of the received signal
1 for the adaptive filters 121 and 123. Like the delay processing circuit 200, the
delay processing circuit 300 processes the output signal of the magnitude processing
circuit 605 to converge the coefficients to correct values. The magnitude compensating
circuit 400 corrects the magnitude of the output signal of the magnitude processing
circuit 606 to compensate the shift of the acoustic image in the acoustic space caused
by the delay processing of the delay processing circuit 300. The delay processing
circuit 300 corrects the magnitude of the output signal of the magnitude processing
circuit 605 whenever the magnitude compensating circuit 400 performs magnitude correction.
Likewise, the delay processing circuit 200, the delay processing circuit 300 and the
magnitude compensating circuit 400 can be constructed in the configuration shown in
Fig. 3, 6, 9, 10, 12 or 13. When the configuration of Fig. 3, 6 or 9 is used, the
filter 213 is modified from that of Fig. 4. When the configuration of Fig. 9, 12 or
13 is used, the filter 230 is modified from that of Fig. 4.
[0061] Fig. 18 is a block diagram illustrating the configuration of the filter 213 containing
the delay processing circuit 300 with the configuration of Fig. 3, 6 or 9 and the
configuration of the filter 230 containing the delay processing circuit 300 with the
configuration of Fig. 9, 12 or 13. Here, it is assumed that the filter is an L-tap
FIR filter. However, the filter may be in another type such as an IIR filter. Unlike
the configuration of Fig. 4, the coefficient multipliers g
1, g
2, ···, g
L-1 are inserted in cascade to the coefficient multipliers c
0, c
1, c
2, ··, c
L-1, (except c
0). This means that the coefficient multipliers c
0, g
1 c
1, .., g
L-1 c
L-1 are equivalently substituted for the coefficient multipliers c
0, c
1, c
2, ···, c
L-1 of Fig. 4. The operation of the circuit of Fig. 18 is equal to that of the circuit
of Fig. 4. Hence, it is apparent that the use of the filters of Fig. 4 allow the coefficient
values c
0, c
1, ···, c
L-1 of the coefficient multipliers 2132
1, 2132
2, ···, 2132
L-1 to be replaced with the coefficient values g
1 c
1, ··, g
L-1 c
L-1.
[0062] Fig. 19 is a block diagram illustrating the configuration of the filter 213 having
the magnitude compensating circuit 400 with the configuration of Fig. 3, 6 or 9 as
well as the configuration of the filter 230 having the magnitude compensating circuit
400 with the configuration of Fig. 9, 12 or 13. Here, it is assumed that the filter
is an L-tap FIR filter. However, the filter may be in another type such as an IIR
filter. Unlike the configuration of Fig. 18, the delay elements 2132
1, 2132
2, ···, 2132
L-1 are not used.
[0063] The filters shown in Figs. 18 and 19 operate complimentarily. That is, the shift
of an acoustic image is corrected by a group of coefficient multipliers 2137i (i =
1, 2, ···, L-1) or a group of coefficient multipliers 2138
i (i = 1, 2, ···, L-1). The two groups correspond to each other. The principle of correcting
a shift of the acoustic image caused by delay variations by magnitude correction is
disclosed in "Medical Research Council Special Report" , No. 166, 1932, pp. 1-32 (reference
11) and "Journal of Acoustical Society of America" , Vol. 32, 1960, pp. 685-692 (reference
12) and "Journal of Acoustical Society of America" , Vol. 94, 1993, pp. 98-110 (reference
13). In the example of Fig. 13, since the output of the delay processing circuit 200
is delayed, the acoustic images of acoustic signals reproduced for the talker 11 by
the loudspeakers 3 and 4 move toward the loudspeaker 3. In order to compensate for
the shift, the magnitude of the signal radiated from the loudspeaker 4 in the acoustic
space is increased while the magnitude of the signal radiated from the loudspeaker
3 in the acoustic space is decreased.
[0064] According to the reference 13, in order to move the acoustic image by magnitude correction,
with the total power of the received signals 1 and 2 maintained constant, the following
equation must be held between the power P
1 (dB) and the power P
2 (dB), the powers of these signals.

where C is a positive constant.
[0065] For the power of the received signal 1 and the power of the received signal 2 which
are P
1 bar dB and P
2 bar dB respectively before the magnitude correction, the following equations must
be satisfied:


where ΔP/2 is a power correction value. For that reason, the coefficients g
i and f
1 of the multipliers for correction corresponding to the filters shown in Figs, 18
and 19 can be determined by the equations (14) and (15).


where ΔP
i is a power compensation coefficient required to compensate the received signal delayed
by
i samples.
[0066] Fig. 20 illustrates another configuration of the filter of Fig. 19. In Fig. 19, a
plurality of coefficient multiplier groups, each having coefficient multipliers serially
connected, are connected in parallel. Referring to Fig. 20, the multiplier groups
are integrated into one coefficient multiplier unit. The input signal is input to
the input terminal 2130.
[0067] The multiplier 2139 with a time varying coefficient
cΣ multiplies the input signal by
cΣ. The resultant product is output via the output terminal 2134.
[0068] cΣ is expressed by the following equation.

[0069] The delay processing circuit 300 applied to the received signal 2 and the magnitude
compensating circuit 400 applied to the received signal 1 have been described by referring
to Figs. 13, and 18 to 20. However, a similar description can be apparently applied
to the delay processing circuit 300 used for the received signal 1 and the magnitude
compensating circuit 400 used for the received signal 2.
[0070] Fig. 21 depicts a multi-channel echo canceller with two channels for the received
signals and two channels for the transmission signals, according to the sixth embodiment
of the present invention. Like the relationship between Figs. 1 and 13, Fig. 21 shows
the configuration where the pre-processing circuit 510 and the non-linear processing
circuit 600 in Fig. 17 are exchanged in the order.
[0071] Fig. 22 depicts a multi-channel echo canceller with two channels for the received
signals and two channels for the transmission signals, according to the seventh embodiment
of the present invention. Fig. 22 shows the configuration where the non-linear processing
circuit 610 is substituted for the non-linear processing circuit 600 in Fig. 17. The
effect is apparent from the relationship between the configuration of Fig. 1 and the
configuration of Fig. 14.
[0072] Fig. 23 depicts a multi-channel echo canceller with two channels for the received
signals and two channels for the transmission signals, according to the eighth embodiment
of the present invention. Fig. 23 shows the configuration where the non-linear processing
circuit 610 and the pre-processing circuit 510 in Fig. 22 are exchanged in the order.
The effect is apparent from the relationship between the configuration of Fig. 14
and the configuration of Fig. 16.
[0073] According to the above-mentioned embodiments, echo cancellation for multi-channel
teleconferencing systems has been discussed. However, a similar discussion can be
applied to single-channel, multi-point teleconferencing systems being another application
of multi-channel echo cancellation. In the single-channel, multi-point teleconferencing
system, an attenuation and a delay are suitably added to the voice of a talker received
by a single microphone such that the talker is localized at a desired position between
the loudspeakers used on the receiver side. A plurality of such signals corresponding
to the number of loudspeakers on the receiving side are created. When the number of
loudspeakers on the receiving side is 2, the two signals, to which the attenuation
process and the delay processing in the prior-art configuration shown in Fig. 24 are
applied, respectively, correspond to the received signal 1 and the received signal
2. Hence, the embodiments of the present invention can be applied without any change.
[0074] Here, an example where the received signal 1 and the received signal 2, the first
mixed signal 14 and the second mixed signal 15 are used has been described. However,
the present invention is applicable to the case where plural received signals and
a single or plural transmission signals are used. In the examples, the received signal
propagates from loudspeakers to a microphone via spatial acoustic paths and the acoustic
echo received by the microphone is removed. However, the present invention is applicable
to cancel echoes other than acoustic echoes, for example, echoes such as crosstalk.
Moreover, examples where finite impulse response filters based on the LMS algorithm
are used as the adaptive filters 121, 122, 123 and 124 have been described. However,
the present invention can use an arbitrary type of adaptive filters. For example,
when the finite impulse response filters based on the NLMS algorithm are used, the
filter coefficient updates are expressed by:


[0075] The sequential regression algorithm (SRA) (reference 8) and the RLS algorithm (reference
9) may also be used as the algorithm for adaptive filters. Infinite impulse response
(IIR) adaptive filters may be used instead of the finite impulse response (FIR) adaptive
filters. Moreover, the subband adaptive filters or transform-domain adaptive filters
may be used.
[0076] In the multi-channel echo canceling system according to the present invention, one
received signal is filtered to create a supplement signal. Then, the adaptive filters
are operated using a new received signal obtained by time-multiplexing the supplement
signal with the original received signal. The adaptive filter operates in response
to a signal obtained by multiplexing the original received signal with a new supplement
signal. Thus, plural adaptive filters are not used to estimate echoes caused by signals
propagating from one signal source via plural paths. Consequently, the number of conditional
equations for adaptive filter coefficients and the problem does not occur that the
solution becomes indefinite. This is because the conventional multi-channel echo canceller
based on linear combination can use only the three conditionals in (7). However, the
present invention can use six conditionals, that is twice the number in the prior
art, so that the adaptive filter coefficients converge to optimum values which are
uniquely determined.
[0077] Moreover, according to the present invention, the multiplex parameter for the original
received signal and the supplement signal is controlled based on the nature of the
received signal. Meanwhile, the sound shift caused by the introduction of the supplement
signal is canceled by correcting the magnitude of the input signals. Thus, the degradation
in sound quality of the audible received signal directly supplied to the loudspeaker
is suppressed, so that good sound quality can be maintained. Moreover, the convergence
time can be shortened by a synergistic effect regarding the non-linear processing
and reduction in correlation between the received signals due to the introduction
of the supplement signal.
[0078] The entire disclosure of Japanese Patent Application No. 10-248025 filed on September
02, 1998 including specification, claims, drawing and summary is incorporated herein
by reference in its entirety.
1. A multi-channel echo canceling method suitable in a system having a plurality of received
signals and a single or plural transmission signals, wherein echoes are removed by
subtracting replicas from mixed signals containing a plurality of echoes and said
transmission signals, said echoes being caused by said plurality of received signals
propagating in spatial acoustic paths or caused by crosstalks on transmission lines,
said replicas corresponding to said plurality of echoes created by plural adaptive
filters which receive said plurality of received signals, comprising the steps of:
selecting one of non-linearly processed signals obtained by non-linearly processing
said plurality of received signals, as a non-linear selected received signal;
filtering said non-linear selected received signal to produce a processed signal;
time-multiplexing said non-linear selected received signal with said processed signal
to produce a new received signal; and
inputting said new received signal and all the remaining non-linearly processed signals
to corresponding adaptive filters respectively, thus, creating plural replicas.
2. The multi-channel echo canceling method defined in Claim 1, further comprising the
step of filtering said non-linear selection received signal using plural time-varying
coefficients with values of zero and values of non-zero, whereby said non-linear selected
received signal and said processed signal are time-multiplexed.
3. The multi-channel echo canceling method defined in Claim 1, wherein the period for
which said non-linear selection received signal and said process signal are time-multiplexed
is constant and longer than the sampling period of the signals to be multiplexed.
4. The multi-channel echo canceling method defined in Claim 1, wherein the period with
which said non-linear selected received signal and said processed signal are time-multiplexed
is longer than the sampling period of the signals to be multiplexed and changes depending
on analysis results of the signals to be multiplexed.
5. A multi-channel echo canceling method suitable in a system having a plurality of received
signals and a single or plural transmission signals, wherein echoes are removed by
subtracting replicas from mixed signals containing a plurality of echoes and said
transmission signals, said echoes being caused by said plurality of received signals
propagating in spatial acoustic paths or caused by crosstalks on transmission lines,
said replicas corresponding to said plurality of echoes created by plural adaptive
filters which receive said plurality of received signals, comprising the steps of:
selecting one of said plurality of received signals, as a selected received signal;
filtering said selected received signal to produce a processed signal;
time-multiplexing said selected received signal with said processed signal to produce
a new received signal;
non-linearly processing said new received signal and all other received signals to
produce non-linearly processed signals; and
inputting said non-linearly processed signals to corresponding adaptive filters respectively,
thus, creating plural replicas.
6. The multi-channel echo canceling method defined in Claim 5, wherein the period with
which said selected received signal and said processed signal are time-multiplexed
is constant and longer than the sampling period of the signals to be multiplexed.
7. The multi-channel echo canceling method defined in Claim 5, wherein the period with
which said selected received signal and said processed signal are time-multiplexed
is longer than the sampling period of the signals to be multiplexed and changes depending
on analysis results of the signals to be multiplexed.
8. A multi-channel echo canceling method suitable in a system having a plurality of received
signals and a single or plural transmission signals, wherein echoes are removed by
subtracting replicas from mixed signals containing a plurality of echoes and said
transmission signals, said echoes being caused by said plurality of received signals
propagating in spatial acoustic paths or caused by crosstalks on transmission lines,
said replicas corresponding to said plurality of echoes created by plural adaptive
filters which receive said plurality of received signals, comprising the steps of:
selecting one of said plurality of received signals to create a selected received
signal;
non-linearly processing said selected received signal as a non-linear selected received
signal;
filtering said non-linear selected received signal to produce a processed signal;
time-multiplexing said non-linear selected received signal with said processed signal
to produce a new received signal;
correcting the magnitude of said new received signal and then inputting the corrected
received signal to a corresponding adaptive filter, instead of said selected received
signal, to create one of said plural replicas;
non-linearly processing received signals other than said selected received signal
to produce non-linearly processed signals and then correcting magnitudes of said non-linearly
processed signals to produce magnitude-corrected signals; and
inputting said magnitude-corrected signals to corresponding adaptive filters respectively,
instead of received signals other than said selected received signal, thus, creating
plural remaining replicas.
9. The multi-channel echo canceling method defined in Claims 1, 5 or 8 wherein said non-linear
processing step comprises the steps of multiplying the received signal in another
channel by a constant according to the polarity of the input signal and then adding
the product to said input signal.
10. The multi-channel echo canceling method defined in Claims 1 or 8, further comprising
the step of selecting said non-linear selected received signal or said processed signal
to implement time-multiplexing.
11. The multi-channel echo canceling method defined in Claim 8, wherein the period for
which said non-linear selected received signal and said processed signal are time-multiplexed
is constant and longer than the sampling period of the signals to be multiplexed.
12. The multi-channel echo canceling method defined in Claim 8, wherein the period with
which said non-linear selected received signal and said processed signal are time-multiplexed
is longer than the sampling period of the signals to be multiplexed and changes depending
on analysis results of the signals to be multiplexed.
13. A multi-channel echo canceling method suitable in a system having a plurality of received
signals and a single or plural transmission signals, wherein echoes are removed by
subtracting replicas from mixed signals containing a plurality of echoes and said
transmission signals, said echoes being caused by said plurality of received signals
propagating in a spatial acoustic paths or caused by crosstalks on transmission lines,
said replicas corresponding to said plurality of echoes created by plural adaptive
filters which receive said plurality of received signals, comprising the steps of:
selecting one of said plurality of received signals, as a selected received signal;
filtering said selected received signal to produce a processed signal;
time-multiplexing said selected received signal with said processed signal to produce
a new received signal;
correcting a magnitude of said new received signal;
non-linearly processing said new received signal to input to a corresponding adaptive
filter,
instead of said selected received signal, thus, creating one of said plural replicas;
correcting magnitudes of received signals other than said selected received signal;
non-linearly processing said corrected received signals to produce non-linear magnitude
corrected signals; and
inputting said non-linear magnitude corrected signals to corresponding adaptive filters,
instead of received signals other than said selected received signal, thus, creating
plural replicas.
14. The multi-channel echo canceling method defined in Claims 1, 5, 9 or 13, wherein said
non-linear processing step comprises the steps of multiplying the received signal
in another channel by a constant according to the polarity of the input signal and
then adding the product to said input signal.
15. The multi-channel echo canceling method defined in Claims 1, 5 or 13, further comprising
the step of selecting said non-linear selected received signal or said selected received
signal or said processed signal to implement time-multiplexing.
16. The multi-channel echo canceling method defined in Claims 5, 8 or 13, further comprising
the step of filtering said selected received signal using plural time-varying coefficients
with values of zero and values of non-zero, whereby said (non-linear) selected received
signal and said processed signal are time-processed.
17. The multi-channel echo canceling method defined in Claim 13 wherein the period with
which either said selected received signal and said processed signal are time-multiplexed
is constant and longer than the sampling period of the signals to be multiplexed.
18. The multi-channel echo canceling method defined in Claim 13 wherein the period with
which either said selected received signal and said processed signal are time-multiplexed
is longer than the sampling period of the signals to be multiplexed and changes depending
on analysis results of the signals to be multiplexed.
19. A multi-channel echo canceling apparatus suitable in a system having a plurality of
received signals and a single or plural transmission signals, wherein echoes are removed
by subtracting replicas from mixed signals containing a plurality of echoes and said
transmission signals, said echoes being caused by said plurality of received signals
propagating in a spatial acoustic paths or caused by crosstalks on transmission lines,
said replicas corresponding to said plurality of echoes created by plural adaptive
filters which receives said plurality of received signals, comprising:
a plurality of adaptive filters each respectively corresponding to said spatial acoustic
path or to said crosstalk path between channels, said adaptive filters creating said
replicas in response to the same received signal as the signal supplied to a corresponding
path;
a non-linear processing circuit for implementing a non-linear process for said plurality
of received signals to generate a plurality of non-linear received signals;
a pre-processing circuit for filtering one of said plurality of non-linear received
signals to generate a processed signal, time-multiplexing said processed signal and
said selected non-linear received signal to be filtered, and then supplying the multiplexed
signal as the new received signal to said plurality of adaptive filters; and
a plurality of subtracters for subtracting said created replicas from said mixed signals;
whereby said plurality of filters are controlled to minimize the outputs of said plurality
of subtracters.
20. The multi-channel echo canceling apparatus defined in Claim 19, wherein said non-linear
processing circuit comprises a plurality of magnitude processing circuits, each of
said magnitude processing circuits including:
a polarity detection circuit for detecting the polarity of the input signal;
a coefficient multiplier for multiplying the input signal in another channel by a
constant;
a multiplier for multiplying the output of said polarity detection circuit by the
output of said coefficient multiplier; and
an adder for adding the output of said multiplier and the input signal;
whereby, according to the polarity of said input signal, said non-linear processing
circuit adds the input signal in another channel multiplied by a constant to the input
signal or subtracts it from the input signal and then outputs the result.
21. The multi-channel echo canceling apparatus defined in Claim 19, wherein said pre-processing
circuit comprises a switch selecting and time-multiplexing said non-linear selected
received signal and said processed signal.
22. The multi-channel echo canceling apparatus defined in Claim 21, wherein said processing
circuit comprises:
a switch for selecting and time-multiplexing said non-linear selected received signal
and said processed signal;
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
and
a time-varying coefficient filter;
whereby the state of said switch is changed with the frequency-divided clock being
the output of said frequency divider to control said time-varying coefficients with
said frequency-divided clock and said reference clock.
23. The multi-channel echo canceling apparatus defined in Claim 21, wherein said pre-processing
circuit comprises:
a switch for selecting and time-multiplexing said non-linear selected received signal
and said processed signal;
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
an analysis circuit for analyzing said selected received signal;
an AND circuit for detecting a coincidence of the frequency-divided clock, being the
output of said frequency divider, and the output of said analysis circuit; and
a time-varying coefficient filter;
whereby the state of said switch is changed with the output of said AND circuit to
control said time-varying coefficients with the output of said AND circuit and said
reference clock.
24. The multi-channel echo canceling apparatus defined in Claim 21, wherein said pre-processing
circuit comprises:
a switch for selecting and time-multiplexing said non-linear selected received signal
and said processed signal;
an analysis circuit for analyzing said non-linear selected received signal; and
a time-varying coefficient filter;
whereby the state of said switch is changed with the output of said analysis circuit
and said time-varying coefficients are controlled with the output of said analysis
circuit and said reference clock.
25. The multi-channel echo canceling apparatus defined in Claim 19, wherein said pre-processing
circuit comprises a time-varying coefficient filter that time-multiplexes said processed
signal and said non-linear received signal by processing said plurality of received
signals with a plurality of time-varying coefficients having zero and non-zero values,
whereby processed signals are produced.
26. The multi-channel echo canceling apparatus defined in Claim 25, wherein said processing
circuit comprises:
a frequency divider for frequency-dividing the reference clock pulses to generate
a new clock having a period longer than the sampling period of the signals to be multiplexed;
and
a time-varying coefficient filter;
whereby said time-varying coefficients are controlled with said frequency-divided
clock, being the output of said frequency divider, and said reference clock.
27. The multi-channel echo canceling apparatus defined in Claim 25, wherein said pre-processing
circuit comprises:
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
an analysis circuit for analyzing said non-linear selected received signal;
an AND circuit for detecting a coincidence of the frequency-divided clock, being the
output of said frequency divider, and the output of said analysis circuit; and
a time-varying coefficient filter;
whereby said time-varying coefficients are controlled by the output of said AND circuit
and said reference clock.
28. A multi-channel echo canceling apparatus suitable in a system having a plurality of
received signals and a single or plural transmission signals, wherein echoes are removed
by subtracting replicas from mixed signals containing a plurality of echoes and said
transmission signals, said echoes being caused by said plurality of received signals
propagating in spatial acoustic paths or caused by crosstalks on transmission lines,
said replicas corresponding to said plurality of echoes created by plural adaptive
filters which receives said plurality of received signals, comprising at least:
a plurality of adaptive filters each respectively corresponding to said spatial acoustic
path or to said crosstalk path between channels, said adaptive filters creating said
replicas in response to the same received signal as the signal supplied to a corresponding
path;
a pre-processing circuit for filtering one of said plurality of received signals to
produce a processed signal and then time-multiplexing said processed signal and the
received signal to be filtered;
a non-linear processing circuit for non-linearly processing the output from said pre-processing
circuit and then supplying the non-linearly processed signals as a new received signal
to plural adaptive filters; and
a plurality of subtracters for subtracting said created replicas from said mixed signal;
whereby said plurality of filters are controlled to minimize the outputs of said plurality
of subtracters.
29. The multi-channel echo canceling apparatus defined in Claim 28, wherein said non-linear
processing circuit comprises a plurality of magnitude processing circuits, each of
said magnitude processing circuits including:
a polarity detection circuit for detecting the polarity of the input signal;
a coefficient multiplier for multiplying the input signal in another channel by a
constant;
a multiplier for multiplying the output of said polarity detection circuit by the
output of said coefficient multiplier; and
an adder for adding the output of said multiplier and the input signal;
whereby, according to the polarity of said input signal, said non-linear processing
circuit adds the input signal in another channel multiplied by a constant to the input
signal or subtracts it from the input signal and then outputs the result.
30. The multi-channel echo canceling apparatus defined in Claim 28, wherein said pre-processing
circuit comprises a switch selecting and time-multiplexing said non-linear selected
received signal and said processed signal.
31. The multi-channel echo canceling apparatus defined in Claim 30, wherein said processing
circuit comprises:
a switch for selecting and time-multiplexing said selected received signal and said
processed signal;
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
and
a time-varying coefficient filter;
whereby the state of said switch is changed with the frequency-divided clock being
the output of said frequency divider to control said time-varying coefficients with
said frequency-divided clock and said reference clock.
32. A multi-channel echo canceling apparatus suitable in a system having a plurality of
received signals and a single or plural transmission signals, wherein echoes are removed
by subtracting replicas from mixed signals containing a plurality of echoes and said
transmission signals, said echoes being caused by said plurality of received signals
propagating in a spatial acoustic paths or caused by crosstalks on transmission lines,
said replicas corresponding to said plurality of echoes created by plural adaptive
filters which receives said plurality of received signals, comprising:
a plurality of adaptive filters each respectively corresponding to said spatial acoustic
path or to said crosstalk path between channels, said adaptive filters creating said
replicas in response to the same received signal as the signal supplied to a corresponding
path;
a non-linear processing circuit for non-linearly processing said plurality of received
signals and then producing a plurality of non-linear processed signals;
a pre-processing circuit for filtering one of said plurality of non-linear processed
signals to produce a process signal, time-multiplexing said processed signal and non-linear
processed signals to be filtered, and then correcting the magnitude of the multiplexed
signals to supply a plurality of adaptive filters as new received signals; and
a plurality of subtracters for subtracting said created replica from said mixed signal;
whereby said plurality of filters are controlled to minimize the outputs of said plurality
of subtracters.
33. The multi-channel echo canceling apparatus defined in Claim 32, wherein said non-linear
processing circuit comprises a plurality of magnitude processing circuits, each of
said magnitude processing circuits including:
a polarity detection circuit for detecting the polarity of the input signal;
a coefficient multiplier for multiplying the input signal in another channel by a
constant;
a multiplier for multiplying the output of said polarity detection circuit by the
output of said coefficient multiplier; and
an adder for adding the output of said multiplier and the input signal;
whereby, according to the polarity of said input signal, said non-linear processing
circuit adds the input signal in another channel multiplied by a constant to the input
signal or subtracts it from the input signal and then outputs the result.
34. The multi-channel echo canceling apparatus defined in Claim 32, wherein said pre-processing
circuit comprises a switch selecting and time-multiplexing said non-linear selected
received signal and said processed signal.
35. The multi-channel echo canceling apparatus defined in Claim 34, wherein said processing
circuit comprises:
a switch for selecting and time-multiplexing said non-linear selected received signal
and said processed signal;
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
and
a time-varying coefficient filter;
whereby the state of said switch is changed with the frequency-divided clock being
the output of said frequency divider to control said time-varying coefficients with
said frequency-divided clock and said reference clock.
36. The multi-channel echo canceling apparatus defined in Claim 34, wherein said pre-processing
circuit comprises:
a switch for selecting and time-multiplexing said non-linear selected received signal
and said processed signal;
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
an analysis circuit for analyzing said selected received signal;
an AND circuit for detecting a coincidence of the frequency-divided clock, being the
output of said frequency divider, and the output of said analysis circuit; and
a time-varying coefficient filter;
whereby the state of said switch is changed with the output of said AND circuit to
control said time-varying coefficients with the output of said AND circuit and said
reference clock.
37. The multi-channel echo canceling apparatus defined in Claim 34, wherein said pre-processing
circuit comprises:
a switch for selecting and time-multiplexing said non-linear selected received signal
and said processed signal;
an analysis circuit for analyzing said non-linear selected received signal; and
a time-varying coefficient filter;
whereby the state of said switch is changed with the output of said analysis circuit
and said time-varying coefficients are controlled with the output of said analysis
circuit and said reference clock.
38. The multi-channel echo canceling apparatus defined in Claim 32, wherein said pre-processing
circuit comprises a time-varying coefficient filter that time-multiplexes said processed
signal and said non-linear received signal by processing said plurality of received
signals with a plurality of time-varying coefficients having zero and non-zero values,
whereby processed signals are produced.
39. The multi-channel echo canceling apparatus defined in Claim 38, wherein said pre-processing
circuit comprises:
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
and
a time-varying coefficient filter;
whereby said time-varying coefficients are controlled with said frequency-divided
clock, being the output of said frequency divider, and said reference clock.
40. The multi-channel echo canceling apparatus defined in Claim 38, wherein said pre-processing
circuit comprises:
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
an analysis circuit for analyzing said non-linear selected received signal;
an AND circuit for detecting a coincidence of the frequency-divided clock, being the
output of said frequency divider, and the output of said analysis circuit; and
a time-varying coefficient filter;
whereby said time-varying coefficients are controlled by the output of said AND circuit
and said reference clock.
41. A multi-channel echo canceling apparatus suitable in a system having a plurality of
received signals and a single or plural transmission signals, wherein echoes are removed
by subtracting replicas from mixed signals containing plural echoes and said transmission
signals, said echoes being caused by said plurality of received signals propagating
in a spatial acoustic paths or caused by crosstalks on transmission lines, said replicas
corresponding to said plurality of echoes created by plural adaptive filters which
receives said plurality of received signals, comprising:
a plurality of adaptive filters each respectively corresponding to said spatial acoustic
path or to said crosstalk path between channels, said adaptive filters creating said
replicas in response to the same received signal as the signal supplied to a corresponding
path;
a pre-processing circuit for filtering one of said plurality of received signals to
produce a processed signal, time-multiplexing said processed signal and the received
signals to be filtered, and then correcting magnitudes of received signals not to
be filtered;
a non-linear processing circuit for non-linearly processing the outputs of said pre-processing
circuit and then supplying non-linear processed signals as new received signals to
a plurality of adaptive filters; and
a plurality of subtracters for subtracting said created replica from said mixed signal;
whereby said plurality of filters are controlled to minimize the outputs of said plurality
of subtracters.
42. The multi-channel canceling apparatus defined in Claim 41, wherein said non-linear
processing circuit comprises a plurality of magnitude processing circuits, each of
said magnitude processing circuits including:
a polarity detection circuit for detecting the polarity of the input signal;
a coefficient multiplier for multiplying the input signal in another channel by a
constant;
a multiplier for multiplying the output of said polarity detection circuit by the
output of said coefficient multiplier; and
an adder for adding the output of said multiplier and the input signal;
whereby, according to the polarity of said input signal, said non-linear processing
circuit adds the input signal in another channel multiplied by a constant to the input
signal or subtracts it from the input signal and then outputs the result.
43. The multi-channel echo canceling apparatus defined in Claim 41, wherein said pre-processing
circuit comprises a switch selecting and time-multiplexing said selected received
signal and said process signal.
44. The multi-channel canceling apparatus defined in Claim 43, wherein said pre-processing
circuit comprises:
a switch for selectively time-multiplexing said selected received signal and said
processed signal;
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
and
a time-varying coefficient filter;
whereby the state of said switch is changed with the frequency-divided clock being
the output of said frequency divider to control said time-varying coefficients with
said frequency-divided clock and said reference clock.
45. The multi-channel echo canceling apparatus defined in Claims 30 or 43,wherein said
pre-processing circuit comprises:
a switch for selecting and time-multiplexing said selected received signal and said
processed signal;
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
an analysis circuit for analyzing said selected received signal;
an AND circuit for detecting a coincidence of the frequency-divided clock, being the
output of said frequency divider, and the output of said analysis circuit; and
a time-varying coefficient filter;
whereby the state of said switch is changed with the output of said AND circuit to
control said time-varying coefficients with the output of said AND circuit and said
reference clock pulses.
46. The multi-channel echo canceling apparatus defined in Claims 30 or 43, wherein said
pre-processing circuit comprises:
a switch for selecting and time-multiplexing said selected received signal and said
processed signal;
an analysis circuit for analyzing said selected received signal; and
a time-varying coefficient filter;
whereby the state of said switch is changed with the output of said analysis circuit
and said time-varying coefficients are controlled with the output of said analysis
circuit and said reference clock.
47. The multi-channel echo canceling apparatus defined in Claims 28 or 41, wherein said
pre-processing circuit comprises a time-varying coefficient filter that time-multiplexes
said process signal and said received signal by processing said plurality of received
signals with a plurality of time-varying coefficients having zero and non-zero values,
whereby processed signals are produced.
48. The multi-channel echo canceling apparatus defined in Claim 47, wherein said processing
of pre-processing circuit comprises:
a frequency divider for frequency-dividing the reference clock to generate a new clock
having a period longer than the sampling period of the signals to be multiplexed;
and a time-varying coefficient filter;
whereby said time-varying coefficients are controlled with said frequency-divided
clock, being the output of said frequency divider, and said reference clock.
49. The multi-channel echo canceling apparatus defined in Claims 28 or 47, wherein said
pre-processing circuit comprising:
a frequency divider for frequency-dividing the reference clock to generate a new clock,
(said signals) having a period longer than the sampling period of the signals to be
multiplexed;
an analysis circuit for analyzing said selected received signal or said non-linear
selected received signal;
an AND circuit for detecting a coincidence of a frequency-divided clock, being the
output of said frequency divider, and the output of said analysis circuit; and
a time-varying coefficient filter;
whereby said time-varying coefficients are controlled by the output of said AND circuit
and said reference clock (pulses).
50. The multi-channel echo canceling apparatus defined in Claims 25 or 47, wherein said
pre-processing circuit comprises:
an analysis circuit for analyzing said selected received signal; and
a time-varying coefficient filter;
whereby said time-varying coefficient is controlled by the output of said analysis
circuit and said reference clock.